# Blooio Inference

> Blooio trains and serves models for business conversations. Bloobability evaluates supplied policies against a conversation and returns probability-based insight.

Source: https://blooio.com/inference

## Summary

Blooio’s inference catalog provides models for evaluating policies against business conversations.

With Bloobability, send a conversation and the policies you want to assess. It estimates how likely each policy holds and whether the conversation can answer it at all. Results can be used for insight on an individual thread or aggregated over time to identify trends.

## Latest: Bloobability v2.0.0

| Family | Version | Status | Released |
|---|---|---|---|
| Bloobability | v2.0.0 | Available | October 8, 2026 |

Two separately trained tiers behind one endpoint. Send any policies you can write down and a conversation, and pick instant or full; for each policy it returns a calibrated probability and how likely the conversation could answer it at all. No roster, nothing to register.

| Tier | `tier` value | Best for |
|---|---|---|
| Full | `full` | The more capable answer, and a signal when facts are missing. |
| Instant | `instant` | The fastest answer, for questions whose facts are stated. |

## Versions

Every release keeps its own page, measurements, and response contract. Deprecated versions keep their page but are no longer served.

| Version | Status | Released | Summary | Model page |
|---|---|---|---|---|
| v2.0.0 (latest) | Available | October 8, 2026 | Two separately trained tiers behind one endpoint: a 4.24B full tier that handles reversals and missing facts, and a 0.76B instant tier for short, explicit questions. | https://blooio.com/inference/bloobability-v2-0-0.md |
| v1.0.0 (previous) | Deprecated | September 2, 2026 | A matcher and a calibrated gauge on one 3.98B Nemotron model, with 79 individually fitted policies. | https://blooio.com/inference/bloobability-v1-0-0.md |

## Release Process

1. **Measured, then published** — Every published figure is read from the release artifacts used to accept the model.
2. **Calibration frozen at release** — The mapping from a decision to a probability is fixed at release, allowing cohorts to be scored on one scale.
3. **Trained, then held out** — Candidates are trained and then scored on data not used to fit or calibrate them. The model page reports missed gates alongside passed ones.
4. **License assembled per release** — Blooio provides a release-specific grant and passes through licenses for components used in the release; upstream terms remain in force.

## Access

Blooio models run behind the same Blooio account and API key as the messaging platform. Every served version keeps its own response contract, so a client pinned to one release keeps reading the same shape. Bloobability v1 is deprecated and no longer available; its endpoint answers 410 and points to `POST /v4/ai/bloobability/v2/decide`.

Access is available through the REST API and MCP.

## Pricing

Free forever with a paid Blooio account.

| Token type | Includes | Current rate |
|---|---|---:|
| Input | Conversation and policies submitted | Free / 1M tokens |
| Output | Returned probabilities and answerable estimates | Free / 1M tokens |

## Links

- HTML page: https://blooio.com/inference
- Bloobability v2.0.0 model page: https://blooio.com/inference/bloobability-v2-0-0.md
- Bloobability v1.0.0 model page: https://blooio.com/inference/bloobability-v1-0-0.md
- Bloobability playground: https://blooio.com/inference/bloobability/playground.md
- Bloobability v1 license: https://blooio.com/inference/bloobability/license.md
- REST API: https://blooio.com/integrations/api.md
- MCP Server: https://blooio.com/integrations/mcp.md
- Pricing: https://blooio.com/pricing.md
- Blooio AI-readable index: https://blooio.com/llms.txt